Full metadata record
DC FieldValueLanguage
dc.contributor.authorChien, Jen-Tzungen_US
dc.contributor.authorTsou, Kai-Weien_US
dc.date.accessioned2019-04-02T06:04:14Z-
dc.date.available2019-04-02T06:04:14Z-
dc.date.issued2018-01-01en_US
dc.identifier.urihttp://hdl.handle.net/11536/150763-
dc.description.abstractThis paper presents a novel memory-augmented neural network for single-channel source separation. We propose a recall neural network (RCNN) where a couple of external memories are realized for sequence-to-sequence learning based on an encoder and a decoder. These memories are learned in a two-pass sensing procedure where the mixed signal is encoded and then decoded (or recalled) as context vectors by using a bidirectional long short-term memory (LSTM) and a LSTM, respectively. These context vectors are integrated in a gating layer. A set of attention weights are calculated to attend the hidden state of decoder to implement a recurrent neural network for source separation. A gated attention mechanism is carried out to fulfill a specialized memory network. The regression errors due to two passes of sensing procedure and one pass of gated attention are jointly minimized to estimate the weight parameters of different components in different layers. The experiments on multi-speaker speech enhancement show that the proposed RCNN consistently outperforms LSTM and neural Turing machine in different settings.en_US
dc.language.isoen_USen_US
dc.subjectlong short-term memoryen_US
dc.subjectsequence-to-sequence learningen_US
dc.subjectrecall neural networken_US
dc.subjectsource separationen_US
dc.titleRECALL NEURAL NETWORK FOR SOURCE SEPARATIONen_US
dc.typeProceedings Paperen_US
dc.identifier.journal2018 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)en_US
dc.citation.spage2956en_US
dc.citation.epage2960en_US
dc.contributor.department電機工程學系zh_TW
dc.contributor.departmentDepartment of Electrical and Computer Engineeringen_US
dc.identifier.wosnumberWOS:000446384603025en_US
dc.citation.woscount0en_US
Appears in Collections:Conferences Paper